DECISIONMAKING

BigData:TheManagement
Revolution
byAndrewMcAfeeandErikBrynjolfsson
FROMTHEOCTOBER2012ISSUE

ou cant manage what you

dont measure.

Theres much wisdom in that saying, which has

been attributed to both W. Edwards Deming and
Peter Drucker, and it explains why the recent
explosion of digital data is so important. Simply
ARTWORK:TAMARCOHEN,HAPPYMOTORING,2010,SILKSCREENON
VINTAGEROADMAP,26X18

put, because of big data, managers can measure,

and hence know, radically more about their

businesses, and directly translate that knowledge into improved decision making and
performance.

Consider retailing. Booksellers in physical stores could always track which books sold and which
did not. If they had a loyalty program, they could tie some of those purchases to individual
customers. And that was about it. Once shopping moved online, though, the understanding of
customers increased dramatically. Online retailers could track not only what customers bought,
but also what else they looked at; how they navigated through the site; how much they were
inuenced by promotions, reviews, and page layouts; and similarities across individuals and
groups. Before long, they developed algorithms to predict what books individual customers
would like to read nextalgorithms that performed better every time the customer responded to
or ignored a recommendation. Traditional retailers simply couldnt access this kind of
information, let alone act on it in a timely manner. Its no wonder that Amazon has put so many
brick-and-mortar bookstores out of business.

The familiarity of the Amazon story almost masks its power. We expect companies that were born
digital to accomplish things that business executives could only dream of a generation ago. But in
fact the use of big data has the potential to transform traditional businesses as well. It may oer
them even greater opportunities for competitive advantage (online businesses have always
known that they were competing on how well they understood their data). As well discuss in
more detail, the big data of this revolution is far more powerful than the analytics that were used
in the past. We can measure and therefore manage more precisely than ever before. We can make
better predictions and smarter decisions. We can target more-eective interventions, and can do
so in areas that so far have been dominated by gut and intuition rather than by data and rigor.

As the tools and philosophies of big data spread, they will change long-standing ideas about the
value of experience, the nature of expertise, and the practice of management. Smart leaders
across industries will see using big data for what it is: a management revolution. But as with any
other major change in business, the challenges of becoming a big dataenabled organization can
be enormous and require hands-onor in some cases hands-oleadership. Nevertheless, its a
transition that executives need to engage with today.

WhatsNewHere?
Business executives sometimes ask us, Isnt big data just another way of saying analytics?
Its true that theyre related: The big data movement, like analytics before it, seeks to glean
intelligence from data and translate that into business advantage. However, there are three key
dierences:

Volume.
As of 2012, about 2.5 exabytes of data are created each day, and that number is doubling every 40
months or so. More data cross the internet every second than were stored in the entire internet
just 20 years ago. This gives companies an opportunity to work with many petabyes of data in a
single data setand not just from the internet. For instance, it is estimated that Walmart collects
more than 2.5 petabytes of data every hour from its customer transactions. A petabyte is one
quadrillion bytes, or the equivalent of about 20 million ling cabinets worth of text. An exabyte
is 1,000 times that amount, or one billion gigabytes.

Velocity.

For many applications, the speed of data creation is even more important than the volume. Realtime or nearly real-time information makes it possible for a company to be much more agile than
its competitors. For instance, our colleague Alex Sandy Pentland and his group at the MIT
Media Lab used location data from mobile phones to infer how many people were in Macys
parking lots on Black Fridaythe start of the Christmas shopping season in the United States.
This made it possible to estimate the retailers sales on that critical day even before Macys itself
had recorded those sales. Rapid insights like that can provide an obvious competitive advantage
to Wall Street analysts and Main Street managers.

Variety.
Big data takes the form of messages, updates, and images posted to social networks; readings
from sensors; GPS signals from cell phones, and more. Many of the most important sources of big
data are relatively new. The huge amounts of information from social networks, for example, are
only as old as the networks themselves; Facebook was launched in 2004, Twitter in 2006. The
same holds for smartphones and the other mobile devices that now provide enormous streams of
data tied to people, activities, and locations. Because these devices are ubiquitous, its easy to
forget that the iPhone was unveiled only ve years ago, and the iPad in 2010. Thus the structured
databases that stored most corporate information until recently are ill suited to storing and
processing big data. At the same time, the steadily declining costs of all the elements of
computingstorage, memory, processing, bandwidth, and so onmean that previously
expensive data-intensive approaches are quickly becoming economical.

As more and more business activity is digitized, new sources of information and ever-cheaper
equipment combine to bring us into a new era: one in which large amounts of digital information
exist on virtually any topic of interest to a business. Mobile phones, online shopping, social
networks, electronic communication, GPS, and instrumented machinery all produce torrents of
data as a by-product of their ordinary operations. Each of us is now a walking data generator. The
data available are often unstructurednot organized in a databaseand unwieldy, but theres a
huge amount of signal in the noise, simply waiting to be released. Analytics brought rigorous
techniques to decision making; big data is at once simpler and more powerful. As Googles
director of research, Peter Norvig, puts it: We dont have better algorithms. We just have more
data.

HowData-DrivenCompaniesPerform

The second question skeptics might pose is this: Wheres the evidence that using big data
intelligently will improve business performance? The business press is rife with anecdotes and
case studies that supposedly demonstrate the value of being data-driven. But the truth, we
realized recently, is that nobody was tackling that question rigorously. To address this
embarrassing gap, we led a team at the MIT Center for Digital Business, working in partnership
with McKinseys business technology oce and with our colleague Lorin Hitt at Wharton and the
MIT doctoral student Heekyung Kim. We set out to test the hypothesis that data-driven
companies would be better performers. We conducted structured interviews with executives at
330 public North American companies about their organizational and technology management
practices, and gathered performance data from their annual reports and independent sources.

Not everyone was embracing data-driven decision making. In fact, we found a broad spectrum of
attitudes and approaches in every industry. But across all the analyses we conducted, one
relationship stood out: The more companies characterized themselves as data-driven, the better
they performed on objective measures of nancial and operational results. In particular,
companies in the top third of their industry in the use of data-driven decision making were, on
average, 5% more productive and 6% more protable than their competitors. This performance
dierence remained robust after accounting for the contributions of labor, capital, purchased
services, and traditional IT investment. It was statistically signicant and economically
important and was reected in measurable increases in stock market valuations.

So how are managers using big data? Lets look

ExpertisefromSurprising
Sources

in detail at two companies that are far from

Oftensomeonecomingfromoutsidean
industrycanspotabetterwaytousebig
datathananinsider,justbecauseso
manynew,unexpectedsourcesofdata
areavailable.Oneofus,Erik,
demonstratedthisinresearchhe
conductedwithLynnWu,nowan
assistantprofessoratWharton.They
usedpubliclyavailablewebsearchdata
topredicthousing-pricechangesin
metropolitanareasacrosstheUnited
States.Theyhadnospecialknowledgeof
thehousingmarketwhentheybegan
theirstudy,buttheyreasonedthat
virtuallyreal-timesearchdatawould

create new businesses, the other to drive more

Silicon Valley upstarts. One uses big data to

sales.

ImprovedAirlineETAs
Minutes matter in airports. So does accurate
information about ight arrival times: If a plane
lands before the ground sta is ready for it, the
passengers and crew are eectively trapped, and
if it shows up later than expected, the sta sits
idle, driving up costs. So when a major U.S.
airline learned from an internal study that about

enablegoodnear-termforecastsabout
thehousingmarketandtheywere
right.Infact,theirpredictionproved
moreaccuratethantheofcialonefrom
theNationalAssociationofRealtors,
whichhaddevelopedafarmorecomplex
modelbutreliedonrelativelyslowchanginghistoricaldata.

10% of the ights into its major hub had at least

Thisishardlytheonlycaseinwhich
simplemodelsandbigdatatrumpmoreelaborateanalyticsapproaches.
ResearchersattheJohnsHopkinsSchool
ofMedicine,forexample,foundthatthey
couldusedatafromGoogleFluTrends(a
free,publiclyavailableaggregatorof
relevantsearchterms)topredictsurges
inu-relatedemergencyroomvisitsa
weekbeforewarningscamefromthe
CentersforDiseaseControl.Similarly,
Twitterupdateswereasaccurateas
ofcialreportsattrackingthespreadof
cholerainHaitiaftertheJanuary2010
earthquake;theywerealsotwoweeks
earlier.

At the time, the airline was relying on the

a 10-minute gap between the estimated time of

arrival and the actual arrival timeand 30% had
a gap of at least ve minutesit decided to take
action.

aviation industrys long-standing practice of

using the ETAs provided by pilots. The pilots
made these estimates during their nal
approach to the airport, when they had many
other demands on their time and attention. In
search of a better solution, the airline turned to
PASSUR Aerospace, a provider of decisionsupport technologies for the aviation industry.
In 2001 PASSUR began oering its own arrival
estimates as a service called RightETA. It
calculated these times by combining publicly
available data about weather, ight schedules,
and other factors with proprietary data the
company itself collected, including feeds from a
network of passive radar stations it had installed

near airports to gather data about every plane in the local sky.

PASSUR started with just a few of these installations, but by 2012 it had more than 155. Every 4.6
seconds it collects a wide range of information about every plane that it sees. This yields a huge
and constant ood of digital data. Whats more, the company keeps all the data it has gathered
over time, so it has an immense body of multidimensional information spanning more than a
decade. This allows sophisticated analysis and pattern matching. RightETA essentially works by
asking itself What happened all the previous times a plane approached this airport under these
conditions? When did it actually land?

After switching to RightETA, the airline virtually eliminated gaps between estimated and actual
arrival times. PASSUR believes that enabling an airline to know when its planes are going to land
and plan accordingly is worth several million dollars a year at each airport. Its a simple formula:

Using big data leads to better predictions, and better predictions yield better decisions.

Speedier,MorePersonalizedPromotions
A couple of years ago, Sears Holdings came to the conclusion that it needed to generate greater
value from the huge amounts of customer, product, and promotion data it collected from its
Sears, Craftsman, and Lands End brands. Obviously, it would be valuable to combine and make
use of all these data to tailor promotions and other oerings to customers, and to personalize the
oers to take advantage of local conditions. Valuable, but dicult: Sears required about eight
weeks to generate personalized promotions, at which point many of them were no longer optimal
for the company. It took so long mainly because the data required for these large-scale analyses
were both voluminous and highly fragmentedhoused in many databases and data
warehouses maintained by the various brands.

In search of a faster, cheaper way to do its analytic work, Sears Holdings turned to the
technologies and practices of big data. As one of its rst steps, it set up a Hadoop cluster. This is
simply a group of inexpensive commodity servers whose activities are coordinated by an
emerging software framework called Hadoop (named after a toy elephant in the household of
Doug Cutting, one of its developers).

Sears started using the cluster to store incoming data from all its brands and to hold data from
existing data warehouses. It then conducted analyses on the cluster directly, avoiding the timeconsuming complexities of pulling data from various sources and combining them so that they
can be analyzed. This change allowed the company to be much faster and more precise with its
promotions. According to the companys CTO, Phil Shelley, the time needed to generate a
comprehensive set of promotions dropped from eight weeks to one, and is still dropping. And
these promotions are of higher quality, because theyre more timely, more granular, and more
personalized. Searss Hadoop cluster stores and processes several petabytes of data at a fraction
of the cost of a comparable standard data warehouse.

Shelley says hes surprised at how easy it has been to transition from old to new approaches to
data management and high-performance analytics. Because skills and knowledge related to new
data technologies were so rare in 2010, when Sears started the transition, it contracted some of
the work to a company called Cloudera. But over time its old guard of IT and analytics
professionals have become comfortable with the new tools and approaches.

The PASSUR and Sears Holding examples illustrate the power of big data, which allows moreaccurate predictions, better decisions, and precise interventions, and can enable these things at
seemingly limitless scale. Weve seen big data used in supply chain management to understand
why a carmakers defect rates in the eld suddenly increased, in customer service to continually
scan and intervene in the health care practices of millions of people, in planning and forecasting
to better anticipate online sales on the basis of a data set of product characteristics, and so on.
Weve seen similar payos in many other industries and functions, from nance to marketing to
hotels and gaming, and from human resource management to machine repair.

Our statistical analysis tells us that what were seeing is not just a few ashy examples but a more
fundamental transformation of the economy. Weve become convinced that almost no sphere of
business activity will remain untouched by this movement.

ANewCultureofDecisionMaking
The technical challenges of using big data are very real. But the managerial challenges are even
greaterstarting with the role of the senior executive team.

MutingtheHiPPOs.
One of the most critical aspects of big data is its impact on how decisions are made and who gets
to make them. When data are scarce, expensive to obtain, or not available in digital form, it
makes sense to let well-placed people make decisions, which they do on the basis of experience
theyve built up and patterns and relationships theyve observed and internalized. Intuition is
the label given to this style of inference and decision making. People state their opinions about
what the future holdswhats going to happen, how well something will work, and so onand
then plan accordingly. (See The True Measures of Success, by Michael J. Mauboussin, in this
issue.)

Bigdataspowerdoesnoterasetheneedfor
visionorhumaninsight.
For particularly important decisions, these people are typically high up in the organization, or
theyre expensive outsiders brought in because of their expertise and track records. Many in the
big data community maintain that companies often make most of their important decisions by
relying on HiPPOthe highest-paid persons opinion.

To be sure, a number of senior executives are genuinely data-driven and willing to override their
own intuition when the data dont agree with it. But we believe that throughout the business
world today, people rely too much on experience and intuition and not enough on data. For our
research we constructed a 5-point composite scale that captured the overall extent to which a
company was data-driven. Fully 32% of our respondents rated their companies at or below 3 on
this scale.

Newroles.
Executives interested in leading a big data transition can start with two simple techniques. First,
they can get in the habit of asking What do the data say? when faced with an important
decision and following up with more-specic questions such as Where did the data come
from?, What kinds of analyses were conducted?, and How condent are we in the results?
(People will get the message quickly if executives develop this discipline.) Second, they can allow
themselves to be overruled by the data; few things are more powerful for changing a decisionmaking culture than seeing a senior executive concede when data have disproved a hunch.

When it comes to knowing which problems to tackle, of course, domain expertise remains
critical. Traditional domain expertsthose deeply familiar with an areaare the ones who know
where the biggest opportunities and challenges lie. PASSUR, for one, is trying to hire as many
people as possible who have extensive knowledge of operations at Americas major airports. They
will be invaluable in helping the company gure out what oerings and markets it should go after
next.

As the big data movement advances, the role of domain experts will shift. Theyll be valued not
for their HiPPO-style answers but because they know what questions to ask. Pablo Picasso might
have been thinking of domain experts when he said, Computers are useless. They can only give
you answers.

FiveManagementChallenges
GettingStarted
YoudontneedtomakeenormousupfrontinvestmentsinITtousebigdata
(unlikeearliergenerationsofIT-enabled
change).Heresoneapproachtobuilding
acapabilityfromthegroundup.

Companies wont reap the full benets of a

transition to using big data unless theyre able to
manage change eectively. Five areas are
particularly important in that process.

Leadership.

1.Pickabusinessunittobethetesting
ground.Itshouldhaveaquant-friendly
leaderbackedupbyateamofdata
scientists.

Companies succeed in the big data era not

2.Challengeeachkeyfunctiontoidentify
vebusinessopportunitiesbasedonbig
data,eachofwhichcouldbeprototyped
withinveweeksbyateamofnomore
thanvepeople.

clear goals, dene what success looks like, and

3.Implementaprocessforinnovation
thatincludesfoursteps:
experimentation,measurement,sharing,
andreplication.
4.KeepinmindJoysLaw:Mostofthe
smartestpeopleworkforsomeoneelse.
Openupsomeofyourdatasetsand
analyticchallengestointerestedparties
acrosstheinternetandaroundthe
world.

simply because they have more or better data,

but because they have leadership teams that set

ask the right questions. Big datas power does

not erase the need for vision or human insight.
On the contrary, we still must have business
leaders who can spot a great opportunity,
understand how a market is developing, think
creatively and propose truly novel oerings,
articulate a compelling vision, persuade people
to embrace it and work hard to realize it, and
deal eectively with customers, employees,
stockholders, and other stakeholders. The
successful companies of the next decade will be
the ones whose leaders can do all that while
changing the way their organizations make
many decisions.

Talentmanagement.
As data become cheaper, the complements to data become more valuable. Some of the most
crucial of these are data scientists and other professionals skilled at working with large quantities
of information. Statistics are important, but many of the key techniques for using big data are
rarely taught in traditional statistics courses. Perhaps even more important are skills in cleaning
and organizing large data sets; the new kinds of data rarely come in structured formats.
Visualization tools and techniques are also increasing in value. Along with the data scientists, a
new generation of computer scientists are bringing to bear techniques for working with very large
data sets. Expertise in the design of experiments can help cross the gap between correlation and
causation. The best data scientists are also comfortable speaking the language of business and
helping leaders reformulate their challenges in ways that big data can tackle. Not surprisingly,
people with these skills are hard to nd and in great demand. (See Data Scientist: The Sexiest
Job of the 21st Century, by Thomas H. Davenport and D.J. Patil, in this issue.)

Technology.

The tools available to handle the volume, velocity, and variety of big data have improved greatly
in recent years. In general, these technologies are not prohibitively expensive, and much of the
software is open source. Hadoop, the most commonly used framework, combines commodity
hardware with open-source software. It takes incoming streams of data and distributes them
onto cheap disks; it also provides tools for analyzing the data. However, these technologies do
require a skill set that is new to most IT departments, which will need to work hard to integrate
all the relevant internal and external sources of data. Although attention to technology isnt
sucient, it is always a necessary component of a big data strategy.

Decisionmaking.
An eective organization puts information and the relevant decision rights in the same location.
In the big data era, information is created and transferred, and expertise is often not where it
used to be. The artful leader will create an organization exible enough to minimize the not
invented here syndrome and maximize cross-functional cooperation. People who understand
the problems need to be brought together with the right data, but also with the people who have
problem-solving techniques that can eectively exploit them.

Companyculture.
The rst question a data-driven organization asks itself is not What do we think? but What do
we know? This requires a move away from acting solely on hunches and instinct. It also requires
breaking a bad habit weve noticed in many organizations: pretending to be more data-driven
than they actually are. Too often, we saw executives who spiced up their reports with lots of data
that supported decisions they had already made using the traditional HiPPO approach. Only
afterward were underlings dispatched to nd the numbers that would justify the
decision.Without question, many barriers to success remain. There are too few data scientists to
go around. The technologies are new and in some cases exotic. Its too easy to mistake correlation
for causation and to nd misleading patterns in the data. The cultural challenges are enormous,
and, of course, privacy concerns are only going to become more signicant. But the underlying
trends, both in the technology and in the business payo, are unmistakable.

The evidence is clear: Data-driven decisions tend to be better decisions. Leaders will either
embrace this fact or be replaced by others who do. In sector after sector, companies that gure
out how to combine domain expertise with data science will pull away from their rivals. We cant
say that all the winners will be harnessing big data to transform decision making. But the data tell
us thats the surest bet.

AversionofthisarticleappearedintheOctober2012issueofHarvardBusinessReview.